A two-step gradient estimation approach for setting supply chain operating parameters

In earlier work, we found retrospective optimization to be effective for setting policy parameters in supply chains with relatively simple structures. This method finds these parameters by solving an integer program over a single randomly generated sample path. Initial efforts to extend this methodology to more complex settings were in many cases too slow to be effective. In response to this, in this research we combine retrospective optimization over a relatively short time horizon with stochastic approximation gradient search algorithms, an approach that proves to be fast and effective. We compare this approach to retrospective optimization without gradient search on simple serial supply chains where the solution is known, and then use it for effective inventory positioning in more complex biopharmaceutical supply chains.